smbd & nmbd in a Cluster environment, single or multiple instances?

[Axel Thimm]

Hello,

we are about to integrate a TruCluster consisting of two machines into our 50+
SAMBA_2_2 (cvs) served OSF1 network (serving 500+ Windows clients).

The question is how to run smbd & nmbd on the clustered machines. nmbd should
IMHO be run as a single-instance process (due to the unshareable WINS and
Browsing lists), but we wonder whether smbd could be run as parallel instances
for load ballancing reasons.

The details are not Tru64 specific: Clusters can be configured to appear as a
single machine to the rest of the net. Some routing interface decides what to
do with incoming IP connections. Processes can be single-instance, meaning
that they run only on one machine at a time, but may be moved to another
cluster machine, if that machine fails. Processes may also be set up to be
parallel running with the router deciding upon load level which member of the
cluster will get the incoming connection.

The last idiom might be suited for smbd. netbios-ssn would be no problem, as
the router knows upon the session id which cluster member had been
chosen. There might be a problem with netbios-dgm. Those might be distributed
across different cluster members. So the question might boil down to "Are
there dependencies between different NetBIOS datagrams or between datagramms
and sessions?" (Of course all members need to offer exactly the same
resources).

As there are other cluster plattforms with similar functionality (e.g. Linux
clusters), I hope that this may be a known issue in this technical mailing
list.

Thanks in advance & Regards, Axel.
-- 
Axel.Thimm at physik.fu-berlin.de